Abstract:Due to the complex production environment of coal mines,the fine-tuning method for fault diagnosis of coal mining machine bearings is prone to loss data.Based on this,a momentum comparison bi-tuning strategy (MCBiT) was proposed,which could fully explore the discriminative knowledge of data labels and the inherent structure of target data.By transforming the 1D vibration signal of the coal mining machine through the Graman angle difference field and inputting it into the MCBiT,two branches were integrated on the pre-trained backbone of the ImageNet to enhance traditional fine-tuning.One classifier with contrastive cross entropy loss was used to better utilize label knowledge,and the other projector with contrastive learning loss was used to mine the intrinsic structure of the data.The proposed method was tested on six publicly available rotating machinery fault diagnosis datasets and compared with other methods.The results show that the proposed momentum comparison tuning method can effectively build a larger and consistent data sample,and it provide support for the improvement of coal mining machine fault prediction accuracy.